It depends. calibre will search through the device on connecting and compile a list of books, extracting the title, author, size, and publisher.
This info gets cached as %KindleDrive%:\metadata.calibre and the cached data gets read first. Any books already cached can be ignored, and only the new stuff gets scanned. This is done for the sake of speed.
Whenever you send a book from calibre, it sets the metadata from calibre's records, and gains a cover, comments, & series info. This is almost always different than the ebook-metadata, if you are using plugboards. I don't know how matches are performed when the book is imported from the device, very possibly it remembers for that session that the two are the same.
Whenever calibre connects, it first reads data from the cache, then indexes any other books on the device, writes THAT to the cache, and then tries to match books. This is the important part:
Matches are made when the title and author are the same. If you have changed the metadata in calibre, the match will fail. The next device will also fail, perhaps because the file knows it's title is ****: A Novel or some such garbage, perhaps because some other change you made.
You can right-click to "Match Book" and tell calibre two books are the same, and it will update the metadata cache with the calibre metadata. Or send the book from calibre. Then you get the new metadata on the reader, too. You will probably have to delete the old copy first, unless you do as I do and use the old filename as the save path, in which case the old file AND metadata gets overwritten. (This is so the .sdr folders match up. I do a bunch of extra configurations.
)